Does a Habit Take 21 Days? When Someone Actually Measured It, the Answer Ranged From 18 to 254
BaroSit · 2026-07-31 · 📝 블로그
"It takes 21 days to form a habit."
You have heard it so often that the source has probably stopped mattering. It shows up every January, and every time someone decides to start going to the gym.
I went looking for where the number came from. Somewhere in the search the question changed on me. What mattered more than the origin was whether anyone had actually measured it.
① 21 is not a measured number
Let me be straight about one thing first. There are various accounts of where this figure came from, and I could not verify any of them against a primary source. I am not going to repeat what I could not check.
What I can point to is this. Researchers have measured how long it takes for a behaviour to start feeling automatic, and the number they got was not 21.
② Measured, it ran from 18 days to 254
A study published in 2010 went at this question directly.
Ninety-six volunteers each picked one behaviour to do every day. Drinking water, eating a piece of fruit, going for a walk. They repeated it in the same context each time, something like "after breakfast," for twelve weeks, and recorded each day how automatic the behaviour felt.
Automaticity scores climb quickly at first, then flatten out and settle. The researchers worked out, for each person, how long it took to reach that plateau.
It ranged from 18 days to 254.
The same researchers later summarised the finding and reported the average as 66 days. Though the average looks fairly useless here. The fastest and slowest participants were 14 times apart.
One more thing, to be fair to the study. Of the 96 volunteers, 82 produced usable data, and the curve was a good fit for 39 of them.
So this study did not show that people generally need about 66 days. If anything it showed that a single number is the wrong shape of answer.
③ Measured a different way, it points somewhere similar
I wanted to know whether this was one study's quirk, so I kept looking.
A randomised trial published in 2021 followed 192 adults aged 18 to 77. Each picked a daily behaviour and logged it for 84 days. Among those who did form the habit, automaticity peaked after a median of 59 days.
That trial turned up something else worth noting. People who attached the behaviour to an existing routine ("after dinner") did no better or worse than people who attached it to a clock time ("7pm"). What predicted automaticity was whether they actually repeated the planned behaviour.
Not what you hook it to, in other words, but whether the hook produces repetition.
Then in 2023 came a study with a completely different approach. Researchers applied machine learning to over 12 million gym visits and over 40 million hospital handwashing events. Actual behavioural records, not self-report.
Their conclusion: contrary to the popular belief in a "magic number" of days, forming the habit of going to the gym typically took months, while handwashing took weeks.
④ Different behaviours take different lengths of time
The findings converge on a point. What sets the timeline is not the calendar but the behaviour itself.
▸ From the 2010 team's own summary — simple actions like drinking a glass of water became automatic faster than more elaborate routines like doing 50 sit-ups
▸ From the 2023 analysis — handwashing took weeks, going to the gym took months
Obvious once said, and yet completely invisible if you are holding on to a single number like 21.
There is a practical reason this distinction matters. It changes how big you make the thing you start with. For the same goal, choosing the lower-threshold version shortens the road to automatic.
⑤ So what changes today
Three things I took away from all this.
▸ Don't judge at three weeks
There is no basis for treating 21 days as a verdict. If three weeks in it still takes effort, that is not a personal failing. That is roughly what the measurements say to expect.
▸ Don't count a missed day as failure
This was my favourite finding in the 2010 study. Missing a single opportunity did not materially impair the habit formation process, and automaticity soon resumed. What breaks the streak is not the missed day. It is deciding the missed day means you failed.
▸ Worry less about the cue, more about the repetition
Hooking it to a routine and hooking it to a time worked about equally well. What counted was whether the cue actually brought the behaviour with it.
The question about duration is worth rephrasing too.
Not "how many days until it sticks," but "is this small enough that I can keep doing it, however many days it takes."
This series usually talks about sitting and posture, and it is the same problem underneath. Knowing which position is easier on you is one thing. Actually shifting a few times a day is another.
That gap is the only place BaroSit really operates. It does not decide anything for you; it just says something when you have forgotten. It cannot do the repeating for you, but it can help.
→ Have a look at barosit.com if you are curious.
An earlier post covered how the format you write a plan in changes whether you act on it. The evidence on sitting time and movement is collected at barosit.com/science.
Sources
• Lally et al., 2010 · Eur J Soc Psychol 40(6):998–1009 — 96 volunteers each repeated a self-chosen behaviour in a consistent context for 12 weeks. 82 provided sufficient data for analysis; the curve model fitted 62 participants, of whom 39 showed a good fit. Time to reach 95% of the automaticity asymptote ranged from 18 to 254 days. Missing one opportunity did not materially affect habit formation. Volunteers, self-report, self-selected behaviours
• Gardner, Lally & Wardle, 2012 · Br J Gen Pract 62(605):664–6 — The authors' own summary of the above, describing a plateau "after an average of 66 days." Automaticity gains resumed soon after a missed performance. Automaticity peaked more quickly for simple actions (drinking water) than for more elaborate routines (50 sit-ups)
• Keller et al., 2021 · Br J Health Psychol 26(3):807–824 — Randomised trial, 192 adults aged 18–77, 84 days. Among participants who successfully formed habits, automaticity peaked after a median of 59 days. No difference between routine-based and time-based cue planning. Repeated plan enactment was the key predictor of automaticity. Limited to an everyday nutrition behaviour
• Buyalskaya et al., 2023 · PNAS 120(17):e2216115120 — Machine learning applied to over 12 million gym attendance records and over 40 million hospital handwashing observations. Contrary to belief in a "magic number," gym habits took months and handwashing weeks. Observational data, with habit operationalised as predictability of behaviour. A correction (PNAS 2023;120(34):e2312763120) revised the gym estimate to 68–78 days, about 2 to 3 months
This article is general information and is not medical advice.